paddle.fluid.contrib.sparsity.utils. get_mask_2d_best ( mat, n, m ) [source]

Generate 2D n:m sparse pattern mask of the input matrix mat to form sparse matrix with maximun L1 norm .This function would pad each dimension of mat by zero to be a multiples of m before mask generation.

2D n:m sparse pattern: At least \(n \times n\) zeros in every \(m \times m\) block under the constraint of at least n zeros for each row and column.

Note: L1 norm of sparse matrix from Best API is greater than or equal to the one from Greedy.

  • mat (nparray) – The input matrix.

  • n (int) – n of n:m sparse pattern.

  • m (int) – m of n:m sparse pattern.


The 1D n:m sparse mask of mat.

Return type



import numpy as np
import paddle.fluid.contrib.sparsity as sparsity

mat = np.array([[2, 8, 9, 9],
                [9, 1, 3, 9],
                [5, 6, 3, 9],
                [2, 4, 6, 9]])
mask_greedy = sparsity.get_mask_2d_greedy(mat, 2, 4)
mask_best = sparsity.get_mask_2d_best(mat, 2, 4)
print("L1 norm of `greedy` sparse matrix", np.multiply(mat, mask_greedy).sum()) # 56
print("L1 norm of `best` sparse matrix", np.multiply(mat, mask_best).sum()) # 61